NASA and IBM release open AI model that outperforms prior lunar analysis tools

NASA and IBM have released an open-source Lunar Foundation Model designed to help scientists analyze decades of lunar observations and support future Artemis exploration. Trained on data from multiple instruments and missions, including NASA’s Lunar Reconnaissance Orbiter, the model can identify features such as craters, volcanic formations and potential ice deposits across large, complex datasets. NASA and IBM say it outperforms widely used image-analysis methods by up to 23 percent in some tasks and can achieve stronger crater detection with less training data. The organizations also released an open lunar dataset combining tens of thousands of maps and images, along with code and tools that researchers can adapt for their own studies. The resources are publicly available through Hugging Face and GitHub, allowing scientists worldwide to investigate lunar geology, resources and potential landing or exploration sites.